How to Add a New Language on the NLP Map:
Building Resources and Tools for Languages with Scarce Resources
Rada Mihalcea
University of North Texas
rada@cs.unt.edu
Abstract
Those of us whose mother tongue is not English or
are curious about applications involving other lan-
guages, often find ourselves in the situation where
the tools we require are not available. According
to recent studies there are about 7200 different lan-
guages spoken worldwide – without including vari-
ations or dialects – out of which very few have auto-
matic language processing tools and machine read-
able resources.
In this tutorial we will show how we can take
advantage of lessons learned from frequently stud-
ied and used languages in NLP, and of the wealth
of information and collaborative efforts mediated by
the World Wide Web. We structure the presentation
around two major themes: mono-lingual and cross-
lingual approaches. Within the mono-lingual area,
we show how to quickly assemble a corpus for sta-
tistical processing, how to obtain a semantic network
using on-line resources – in particular Wikipedia –
and how to obtain automatically annotated corpora
for a variety of applications. The cross-lingual half
of the tutorial shows how to build upon NLP meth-
ods and resources for other languages, and adapt
them for a new language. We will review automatic
construction of parallel corpora, projecting annota-
tions from one side of the parallel corpus to the
other, building language models, and finally we will
look at how all these can come together in higher-
end applications such as machine translation and
cross-language information retrieval.
938
Vivi Nastase
EML Research gGmbH
Vivi.Nastase@eml-r.villa-bosch.de
Biographies
Rada Mihalcea is an Assistant Professor of Com-
puter Science at the University of North Texas. Her
research interests are in lexical semantics, multi-
lingual natural language processing, minimally su-
pervised natural language learning, and graph-based
algorithms for natural language processing. She
serves on the editorial board of the Journal of Com-
putational Linguistics, the Journal of Language Re-
sources and Evaluations, the Journal of Natural Lan-
guage Engineering, the Journal of Research in Lan-
guage in Computation, and the recently established
Journal of Interesting Negative Results in Natural
Language Processing and Machine Learning.
Vivi Nastase is a post-doctoral fellow at EML Re-
search gGmbH, Heidelberg, Germany. Her research
interests are in lexical semantics, semantic relations,
knowledge extraction, multi-document summariza-
tion, graph-based algorithms for natural language
processing, multilingual natural language process-
ing. She is a co-founder of the Journal of Interest-
ing Negative Results in Natural Language Process-
ing and Machine Learning.